The intelligent site and the intelligent brains behind
With new technologies and 5G on the horizon, we stand before a complexity explosion in telecommunications networks – a complexity that is out of reach for humans alone to deal with. And as the complexity increases we need strategies to handle this, and machine intelligence will be crucial.
Machine intelligence refers to all technologies – including machine learning (ML) and artificial intelligence (AI) – that make machines intelligent enough to solve complex problems without a predefined set of rules for a specific case.
In our network operations, we are enabling intelligent operations by introducing AI/ML-driven automation to predict, prevent and handle events without human intervention. With this, we can reveal insights into network performance and operations that were not previously possible.
This means that we can add value by removing faults at the source, automatically evaluate them, sort, prioritize and then resolve them. Individual sites are now able to proactively manage their health while providing network operations center (NOC) teams with insights into issues that might arise, their impact if nothing is done and what preventative actions should be taken.
At the same time, we can improve customer experience as well as network quality and responsiveness to changing customer needs, by understanding and handling what really matters to customers in an automated and intelligent fashion. We also free up our engineers from having to undertake repetitive manual tasks, allowing them to focus on providing more value-added services.
In short, we are going from event-driven, reactive operations to proactive, data, AI/ML-driven and managed operations.
But how do you build in this level of machine intelligence and who are the intelligent brains behind?
This intelligent solution has been the fruit of effort of several people at Ericsson Research. This is their view of the technology challenges and how we can take control of the complexity.
Building a platform for developing, testing and publishing machine learning models
A site is a very rich source of data since it is tasked with handling all phone calls made by different devices, as well as all data connections from streaming videos to sensor information produced by small IoT devices. As such, it serves a very important role and it cannot afford any down time. But what kind of information generated by this site is so vitally crucial to its own well-being?
To answer this question, and for the purposes of experimentation but also for creating a solution that can help our engineers, we embarked on developing a living cloud-based setup responsible for maintaining all the different machine learning algorithms developed in this project, both in terms of training new models but also for running them efficiently. On top of that, the setup needed to be fault-tolerant and also capable of handling large volumes of data.
As soon as a model reached its maturity it became available to our engineers along with mechanisms that can verify its performance. Under-performing models are sent back to the drawing board to be discarded or redesigned. The beauty of data-driven projects is that you can always find new pieces of information that can help you create a more accurate model.
The Ericsson expert
Konstantinos Vandikas is a Master Researcher at Ericsson Research, currently working in the area of Machine Intelligence on cloud and edge-based deployments. His background is in service-oriented architectures and distributed systems. He has been with Ericsson Research since 2007, actively evolving research concepts from inception to productification. Konstantinos holds a Ph.D. in Computer Science from RWTH Aachen, Germany. He has 12 granted patents and has filed over 50 patent applications. Konstantinos has also co-authored more than 20 scientific publications.
Anomaly detection using machine learning
Millions of different alarms occur at base stations every day based on manually pre-programmed rules and thresholds, and site engineers in the operations have to cope with these enormous amounts of alarms, tickets, and work order management. It is highly difficult for site engineers to understand the complexity of site issues, prioritize them, and understand the hidden patterns and correlations amongst them. Therefore, it becomes vital to detect the abnormal patterns within the complex network operations and to help domain experts see the hidden insights that were previously impossible to see. One study we conducted was base station-based anomaly detection depending on the alarms and tickets, in order to capture unusual site behavior of base stations over time. The lifetime similarity of a particular site with other sites within a mobile operator are revealed very fast by using distributed data analytics tools such as Spark. Creating these insights with the detection of those anomalies help to prioritize the ones that are more important than others.
We work towards enabling Machine Intelligence on base station operations via advanced supervised and unsupervised machine learning models. We study these models to not only prevent the faults from happening by early predictions and proactive decisions, but also to make base stations immune to any imaginative problems that have never occurred before but may happen in the future, using generative models. In parallel, this helps domain experts to understand unseen insights within the complexity of the network issues.
It is also important for an operator to get a larger view of the performance of the network. We have therefore developed methods to find anomalies in patterns for aggregated alarms in large areas of the network and highlight them for the operator. Another method we have developed is comparing different base station sites in the network to find similarities in the alarm patterns. By using these methods, the operator can investigate possible causes for the unusual patterns, such as weather-related issues or misconfigurations in larger parts of the network.
The Ericsson Experts
Selim Ickin, Ph.D. is a Senior Researcher in the Machine Intelligence department at Ericsson Research, Stockholm. His research interest is machine learning and intelligent software-prototyping towards productization. He has worked in numerous data driven machine-learning projects in diverse domains targeting to improve network-based mobile application performance; to reduce subscriber churn rate for a video service provider; and to reduce network operations cost in managed services since when he joined Ericsson Research in late 2014.
Christofer Flinta is a Senior Researcher in the Research Area Machine Intelligence and Automation at Ericsson Research. Since he joined Ericsson in 1999, he has worked in the diverse areas of mobile positioning, VPN, home gateways, IPv6, network-performance measurements, cloud management and machine learning. His current focus is anomaly detection in radio networks and improving cloud performance using machine learning methods. Christofer holds 19 granted patents and has co-authored 2 IETF and ITU-T standards as well as several research papers. He holds a Licentiate degree in Theoretical Physics, BioPhysics, at KTH, Sweden.
Federating radio site information
It is evident that having timely access to relevant data about the network is essential for Network Operations Centre (NOC) personnel to perform their management tasks. The information describing a mobile network radio site comprises basic elements such as site name, location, sector configuration and spectrum allocation, as well as vendor specific configuration of components including baseband, radio units, antenna, mast, main and backup power systems, etc.
Today, to get a complete picture a front office engineer must consult multiple systems, login to them, perform a search and manually do the corresponding correlations, which can be time consuming and may directly affect service availability. Part of the work to make site operations intelligent involved creating site profiles: a sort of digital passport of a site that would contain aggregated description along with the output from machine learning methods and third-party data such as weather.
We borrowed ideas of semantic web to create site federated models using linked data principles. The result is both human- and machine-friendly instant access to all relevant site information that may be needed during site maintenance by NOC engineers and by machine intelligence algorithms.
The Ericsson Expert
Leonid Mokrushin is a senior specialist in cognitive technologies at Ericsson Research. His current focus is on investigating new opportunities within AI in the context of industrial and telco use cases. He joined Ericsson Research in 2007 after postgraduate studies at Uppsala University, Sweden, with a background in real-time systems. He received an M.Sc. in software engineering from Peter the Great St. Petersburg Polytechnic University, Russia in 2001.
Learn more about machine intelligence
Interested in learning more about how our machine intelligence uses machine learning and artificial intelligence to drive systems for automation and network evolution? Discover more on our site and explore how Machine Intelligence differentiates Ericsson’s portfolio and services.